Friday, 1 November 2013

Benchmark that, sucker!

Back in the day of the not-a-law Moore's Law, I had a rule that I only upgraded my Mac when the new system offered three times the performance of the old. That was easy to achieve when moving from, say, a dreadful Power Mac 4400/160 to a Quicksilver G4/933, but it's become progressively more challenging.

My reasoning behind the factor-three rule was straightforward:

  • Most things happen effectively instantaneously, and they aren't going to get any more instantaneous with a faster system.
  • Things that happen merely 'very quickly' add up, but memory capacity and disk performance tend to have more impact here than CPU performance.
  • Things that take 10 minutes are an absolute pain, and are to be avoided. Happily, there never seem to be very many operations that take 10 minutes.
  • Things that take 20 minutes are 'coffee breaks'. When I'm working, I actually want some operations to fall into this category. They can be a pain, but I can also schedule around them during the day.
  • Things that take an hour are where problems start. An hour is too long a break during the working day, which means hour+ processes are best queued up and run overnight.
A factor-three increase in performance starts to move 'overnight' processes into 'coffee break' processes. That's huge. No longer do you have to push export and compress jobs out overnight, you can run them today. Even several times during the day. Processes that used to be marginal for even overnight render can now be batched together, with multiple jobs run in one end-of-day break.

Now, my current edit system is a 2008 8-core Mac Pro, which despite being five years old continues to chug along at maybe half the speed of the fastest current Macs. That's both commendable longevity and frustratingly slow progress. For an example of what I mean: every editor's favourite plugin author Alex Gollner has a newly-minted FCPX benchmark: head to this thread for the file, example results, and discussion.

My Pro takes about 120 seconds, which to be sure ranks as 'getting a bit old', but even the fastest current Mac struggles to reach a factor two improvement. That's good, but hardly represents the improvement one would like to see after half a decade.

This is why the new Mac Pro is interesting. My 2008 model holds its own because I rammed an upgraded video card into it. It's nothing special, but neither are the cards in the current iMacs. The trashcan Pro not only packs a newer graphics card, it packs two of them. Interesting.

Want an example? That G4/933 could handle noise reduction in video, but only for simplistic algorithms and only for an overnight job. For a short film, the 2008 Pro does a better job over a coffee break. The Trashcan? I'll be interested to see how much it can do in real time.

Factors of three turn special-case tools into everyday tools. They also enable completely new tools, and it's been a while since we've had the capacity to handle those on the desktop.

No comments:

Post a Comment